Convolutional Neural Networks
Convolution Operation
Padding
As padding size increases, convolution gets more information, especially at the edges.
The attributes of padding:
- Enable the acquisition of more information.
- Increase the resolution of the output.
Stride
Stride refers to the number of steps the filter moves each time.
The attributes of stride:
- Allow the compression of partial information.
- Reduce the resolution of the output.
Pooling
Pooling is a widely used technique that mainly focuses on reducing the resolution of the output.
The Calculation of Output Size
Define:
* image size:
Then the output size: